Another restatement of Brandolini’s Law. The cost of parroting this kind of information is very low, while the cost of refuting it is very high. And the value an outlet can extract from its readership to fund that refutation is nowhere close to cover its outlay. Maybe a counter is the occasional take-down article can sometimes go more viral than the original claim, but chasing those is probably unprofitable too.
One thought is once it can touch the underlying system, it can provision resources, spawn processes, and persist itself, crossing the line from tool to autonomous entity. I admit you could do that in a browser shell nowadays, just maybe with more restrictions and guardrails.
I don’t have any strong opinions here, but I do think a lower cost to escape the walled gardens agi starts in will be a factor
> It is not yet optimized for desktop OS-level control
Alas, AGI is not yet here. But I feel like if this OS-level of control was good enough, and the cost of the LLM in the loop wasn't bad, maybe that would be enough to kick start something akin to AGI.
Officers are a limited resource, so their deployment matters. Are they assigned to areas that most benefit citizens, or those that most benefit the city? Is the focus on maximizing ticket revenue, or addressing the most dangerous violations, like blocked bike lanes? Are they primarily a revenue tool, a public safety measure, or just extra eyes on the street? Do wealthier neighborhoods receive more enforcement, effectively buying themselves safer streets? Basically I'm wondering does parking enforcement benefit SF residents uniformly?
Officer 0336 is raking it in for the city. I wonder if there is a correlation with the areas which generate a lot of tickets and other city datasets. Perhaps crime rate or average household income?
I think we’ve always shaped ourselves based what we’re capable of building. Think of how infrastructure such as buildings and roadways shape our lives within them. What I do agree with you, is how LLMs are shaping our mental thought how we are offloading a lot of our mental capacities with blind trust in the LLM output.
I agree that these visuals rarely drive decisions on their own. They’re more like supporting tools… useful for framing an argument or guiding a narrative in a presentation, especially when static.
Unless it’s interactive or tied to live data the usefulness of the visualization produced is limited.. it’s a shame and it’s something this tools should pivot towards
> Once concepts are selected, Sphere walks the knowledge graph and generates SQL queries to retrieve data from the warehouse, no manual joins or technical mediation required.
If I had to guess this is how eng pitched it to the business to carve out the time to build this tooling. As with all these internally built schemas, ui’s, tooling, etc… they’re never gonna post how much this is actually used relative to the work arounds ds and eng use in their day to day.
Maybe some reverse attribution to determine how much credit to give different sources in the old top k results. Maybe a llm that is trained on fact finding, and giving proper % of credit. But this would probably require a collective contract in place to benefit site owners, which means its never going to happen.
> these tools are not only useful for writing the final code
This sparked a thought in how a large part of the job is often the work needed to demonstrate impact. I think this aspect is often overlooked by some of the good engineers not yet taking advantage of the AI tooling. LLM loops may not yet be good enough to produce shippable code by themselves, but they sure are capable to help reduce the overhead of these up and out communicative tasks.
“‘It’s still learning’ is a misnomer. The model isn’t learning—we are. LLMs are static after training; all improvement comes from human iteration in the outer loop: fine-tuning, prompt engineering, tool integration, retrieval. Until the outer loop itself becomes autonomous and self-improving, we’re nowhere near AGI. Current hype confuses capability with agency.
Was going to point this out too. One suggestion would be to try this on libraries having recent major semvar bumps. See if the compressed docs do better on the backwards incompatible changes.
Only at the very end does the article call out there is actually a performance aspect if you use panic and recover as intended.
> So it seems that panic and recover can be beneficial to performance in at least some situations.
Namely top level & centralized recovery handling. I'll also point out its important your panic recovery must happen in a deferred function if your kicking off new goroutines. For example, your server library probably has default panic recovery on the goroutines handling inbound requests, but any new goroutines you create as part of handling the request (e.g. to parallelize work), will not have this handling built in. See https://go.dev/blog/defer-panic-and-recover for more.
We’re seeing a clear divide where both competitive gamers and hackers are retreating into their own ecosystems, away from public matchmaking. Public matchmaking has simply become too optimized/lucrative to sustain trust or meaningful competition.
Private matchmaking and closed communities are thriving, raising the average skill ceiling in competitive. Similarly, hacking communities are evolving with easier forms of payment and distribution. The monetary aspects are huge. But most importantly, both cultures push each away. Your persona of someone who plays with integrity and crosses the competitive and hacker mentality is pretty much gone.
I vividly remember the difference intense hockey conditioning camps, sleep and recovery can cause. During these camp I pushed my heart, with many drills being 30+ seconds at max heart rate. Afterwards I was exhausted, chest felt terrible. But I was so tired, I didn’t have the usual movements at night, and had much longer deep sleep. The recovery the next day was so dramatic compared to prior workout sessions. Way less inflammation across my joints. As compared to sessions where I only pushed my heart rate to 30-60%
I’ll add that even if the papers we all wanted were more freely accesible, the replication and completeness of their described methods would be another source of slowdown.
> Interestingly enough, it’s actually more efficient to send text as images: A 512x512 image with a small but readable font can easily fit 400-500 tokens worth of text, yet you’re only charged for 170 input tokens plus the 85 for the ‘master thumbnail’ for a grand total of 255 tokens—far less than the number of words on the image.
Sounds like an arbitrage opportunity for all those gpt wrappers. Price your cost per token the same, send over the prompt via image, pocket the difference?
For most non-perishables it’s the most time efficient in minimizing the number of trips. Also given how often I go, ~once a month, each time I get the same sort of novelty as a Trader Joe's for they will have swapped out many items.
If the “Lindy effect” holds true for the 911, it might be quite a long time before the 911 and its variants go out of style.
The same cannot be said for the cyber truck, and all of the other short lived prototype cars that have very harsh angular lines.
Yeah, this author sounds more of an up-and-coming software developer. The cited example is something that someone should be easily able to do in their day to day language.
The ability for chatGPT to essentially translate/expand your knowledge out to other languages (eg “how do I read in a file to utf8 again” ). It’s all just more leverage and power to solve the mundane, faster. The key point is you have to know the rough solution space already.
For the example the op gave, probably not for an adult. But younger kids might. Also for complex domains, I find that dreams enumerate more possibilities/scenarios than I would have if awake.
I hope there will be a HBR or some other case study done on Yellow, because the interplay between bad loans + union demands + poor management seems like something US businesses need to be more prepared for.
There’s really only gonna be two camps of developers, working on native apps. Folks at some existing large company, with some sort of Apple partnership to cover cost, or folks working on their side projects to scratch some personal itch. There really isn't gonna be anyone in the middle burning money for a nonexistent user base.
One could argue we do already via code generation when we define protobuf definitions or other idl’s. But yeah a chat oriented idl which then code generates c, Go, python, just based on the required problem domain is an interesting vision.
Yep. This is a quintessential example of “I’m going to learn by recreating the essence of a complex thing”… taken to the end. And then a logo is slapped on top. I’m all for js libs but this is not performant. Just use redis.